Weighting of the pooled estimates was performed with the random effects model since it was expected that one or more non-RCTs would be included in the analysis [27],[28] and because this model takes into account the heterogeneity of the data [27].
Cross-sectional reliability estimate were performed, with a Cronbach's α between 0.7-0.95 considered good [ 36].
This estimate is performed with the Zeno method implemented in UltraScan-SOMO (see section SI-1, Supporting Information, for an explanation of the method).
Results are reported as β coefficients with 95% CI, except for serum total IgE levels where inverse transformations of model estimates were performed to provide proportional differences in % with 95% CI.
Sequences were edited using the program Sequencher version 4.7 (GeneCodes, Ann Arbor, Michigan, USA), aligned in Geneious, and estimates of variability were performed with MEGA5 (Tamura et al., 2011).
Estimates of Ne were performed with the linkage disequilibrium method (Hill 1981; Waples 2006) updated for missing data following Peel et al.
Five independent runs for each K – the number of inferred groups estimated by Structure – were performed, with K values ranging from 1 to 15.
To evaluation simultaneously the influence of HIV-1 status and, among those HIV-1-seropositive, time-varying measures of immune suppression, generalized estimating equations (GEE) were performed with a Poisson link, an exchangeable correlation structure, and log time) as an offset [ 21].
Both parametric (constant or exponential model) and non-parametric (BSP) estimates of demographic history were performed with BEAST version 1.4.1 by running one MCMC for 100,000,000 generations with sampling every 10,000th generation.
